Skip to content

M
mtbookv2
Commit 9681d94f by 曹润柘
Options

合并分支 'caorunzhe' 到 'master' 

minor update

查看合并请求 !549
parents 3eb4b5d2 a3fabb8d
正在显示 包含 221 行增加 和 318 行删除
Chapter16/Figures/figure-application-process-of-back-translation.tex
......@@ -41,7 +41,7 @@
\node [anchor=west,fill=red!20,inner sep=0.1em,minimum width=3em,draw=black,line width=0.6pt,rounded corners=2pt](node4-1) at ([xshift=2.0em,yshift=1.6em]node3-2.east){\scriptsize{英语}};
\node [anchor=north,fill=green!20,inner sep=0.1em,minimum width=3em,draw=black,line width=0.6pt,rounded corners=2pt](node4-2) at (node4-1.south){\scriptsize{英语}};
\node [anchor=west,fill=green!20,inner sep=0.1em,minimum width=3em,draw=black,line width=0.6pt,rounded corners=2pt](node4-3) at (node4-1.east){\scriptsize{汉语}};
\node [anchor=west,fill=yellow!20,inner sep=0.1em,minimum width=3em,draw=black,line width=0.6pt,rounded corners=2pt](node4-3) at (node4-1.east){\scriptsize{汉语}};
\node [anchor=north,fill=green!20,inner sep=0.1em,minimum width=3em,draw=black,line width=0.6pt,rounded corners=2pt](node4-4) at (node4-3.south){\scriptsize{汉语}};
......
Chapter16/Figures/figure-bilingual-dictionary-Induction.tex 0 → 100644
\begin{tikzpicture}
%%%%%%%%词典推断------------------------------------------------------------
\begin{scope}
\draw [-,ublue,line width=0.5pt] (0,0)..controls (0.3,0.2) and (0.5,0)..(0.7,-0.2)..controls (0.8,-0.3) and (0.9,-0.4)..(1.1,-0.4)..controls (1.3,-0.4) and (1.3,-0.1)..(1.28,0)..controls (1.26,0.1) and (1.25,0.2)..(1.2,0.3)..controls (1.15,0.4)and (1.2,0.5)..(1.6,0.55)..controls (1.7,0.56) and (1.78,0.5)..(1.85,0.35)..controls (2.0,0.0) and (2.05,-0.1)..(2.05,-0.5)..controls (2.04,-1.1) and (1.5,-1.1)..(0.6,-0.78)..controls (0.5,-0.74) and (0.4,-0.7)..(0.2,-0.5)..controls(0.1,-0.4) and (-0.15,-0.1)..(0,0) ;
\draw [-,red!70,line width=0.5pt] (0.04,-0.5) .. controls (0,-0.4) and (0.4,-0.1)..(0.7,-0.3)..controls (0.9,-0.45) and (1.1,-0.4)..(1.2,-0.3)..controls (1.3,-0.2) and (1.2,0.1).. (1.0,0.3)..controls (0.8,0.5) and (1.0,0.6)..(1.2,0.67)..controls (1.5,0.78) and (1.8,0.5)..(1.9,0.2)..controls(2.1,-0.3) and (2,-0.5)..(1.8,-0.75)..controls (1.5,-1.1) and (1.2,-1.0)..(0.4,-0.8)..controls (0.3,-0.77) and (0.14,-0.755)..(0.04,-0.5);
\draw [-,thick] (-0.7,1.0)--(-0.7,-1.0);
\node [anchor=center](c1) at (-0.1,0){\tiny{$\mathbi{Y}$}};
\node [anchor=center](c2) at (-0.3,-0.7){\tiny{$\mathbi{W}\cdot \mathbi{X}$}};
\node [anchor=center,red!70](cr1) at (0.65,-0.65){\scriptsize{$\bullet$}};
\node [anchor=center,ublue](cb1) at (0.6,-0.5){\scriptsize{$\bullet$}};
\node [anchor=center,red!70](cr2) at (1.65,-0.65){\scriptsize{$\bullet$}};
\node [anchor=center,ublue](cb2) at (1.55,-0.8){\scriptsize{$\bullet$}};
\node [anchor=center,red!70](cr3) at (1.5,0.1){\scriptsize{$\bullet$}};
\node [anchor=center,ublue](cb3) at (1.6,-0.05){\scriptsize{$\bullet$}};
\draw [-,red](0.65,-0.65)--(0.60,-0.62)--(0.66,-0.58)--(0.6,-0.55)--(0.63,-0.52)--(0.6,-0.5);
\draw [-,red](1.65,-0.65)--(1.60,-0.68)--(1.64,-0.72)--(1.56,-0.72)--(1.60,-0.76)--(1.55,-0.8);
\draw [-,red](1.5,0.1)--(1.53,0.08)--(1.49,0.04)--(1.58,0.03)--(1.54,-0.01)--(1.6,-0.05);
\end{scope}
%%%%%%%%X映射到Y空间------------------------------------------------------------
\begin{scope}[xshift=-8.0em]
\draw [-,ublue,line width=0.5pt] (0,0)..controls (0.3,0.2) and (0.5,0)..(0.7,-0.2)..controls (0.8,-0.3) and (0.9,-0.4)..(1.1,-0.4)..controls (1.3,-0.4) and (1.3,-0.1)..(1.28,0)..controls (1.26,0.1) and (1.25,0.2)..(1.2,0.3)..controls (1.15,0.4)and (1.2,0.5)..(1.6,0.55)..controls (1.7,0.56) and (1.78,0.5)..(1.85,0.35)..controls (2.0,0.0) and (2.05,-0.1)..(2.05,-0.5)..controls (2.04,-1.1) and (1.5,-1.1)..(0.6,-0.78)..controls (0.5,-0.74) and (0.4,-0.7)..(0.2,-0.5)..controls(0.1,-0.4) and (-0.15,-0.1)..(0,0) ;
\draw [-,red!70,line width=0.5pt] (0.04,-0.5) .. controls (0,-0.4) and (0.4,-0.1)..(0.7,-0.3)..controls (0.9,-0.45) and (1.1,-0.4)..(1.2,-0.3)..controls (1.3,-0.2) and (1.2,0.1).. (1.0,0.3)..controls (0.8,0.5) and (1.0,0.6)..(1.2,0.67)..controls (1.5,0.78) and (1.8,0.5)..(1.9,0.2)..controls(2.1,-0.3) and (2,-0.5)..(1.8,-0.75)..controls (1.5,-1.1) and (1.2,-1.0)..(0.4,-0.8)..controls (0.3,-0.77) and (0.14,-0.755)..(0.04,-0.5);
\draw [-,thick] (-0.7,1.0)--(-0.7,-1.0);
\node [anchor=center](c1) at (-0.1,0){\tiny{$\mathbi{Y}$}};
\node [anchor=center](c2) at (-0.3,-0.7){\tiny{$\mathbi{W}\cdot \mathbi{X}$}};
\node [anchor=center,red!70](cr1) at (0.65,-0.65){\scriptsize{$\bullet$}};
\node [anchor=center,ublue](cb1) at (0.6,-0.5){\scriptsize{$\bullet$}};
\node [anchor=center,red!70](cr2) at (1.65,-0.65){\scriptsize{$\bullet$}};
\node [anchor=center,ublue](cb2) at (1.55,-0.8){\scriptsize{$\bullet$}};
\node [anchor=center,red!70](cr3) at (1.5,0.1){\scriptsize{$\bullet$}};
\node [anchor=center,ublue](cb3) at (1.6,-0.05){\scriptsize{$\bullet$}};
%%%%%%一堆红色的球
\node [anchor=center,red!70](cr4) at (0.15,-0.6){\Large{$\cdot$}};
\node [anchor=center,red!70](cr5) at (0.3,-0.6){\Large{$\cdot$}};
\node [anchor=center,red!70](cr6) at (0.5,-0.55){\Large{$\cdot$}};
\node [anchor=center,red!70](cr7) at (0.35,-0.4){\Large{$\cdot$}};
\node [anchor=center,red!70](cr8) at (0.4,-0.7){\Large{$\cdot$}};
\node [anchor=center,red!70](cr8) at (0.55,-0.8){\Large{$\cdot$}};
\node [anchor=center,red!70](cr9) at (0.9,-0.8){\Large{$\cdot$}};
\node [anchor=center,red!70](cr10) at (0.9,-0.5){\Large{$\cdot$}};
\node [anchor=center,red!70](cr11) at (1.4,-0.8){\Large{$\cdot$}};
\node [anchor=center,red!70](cr12) at (1.45,-0.3){\Large{$\cdot$}};
\node [anchor=center,red!70](cr13) at (1.35,0.3){\Large{$\cdot$}};
\node [anchor=center,red!70](cr14) at (1.2,0.4){\Large{$\cdot$}};
\node [anchor=center,red!70](cr15) at (1.6,0.45){\Large{$\cdot$}};
%%%%%%一堆蓝色的球
\node [anchor=center,ublue](cb4) at (0.1,-0.2){\Large{$\cdot$}};
\node [anchor=center,ublue](cb5) at (0.3,-0.2){\Large{$\cdot$}};
\node [anchor=center,ublue](cb6) at (0.5,-0.25){\Large{$\cdot$}};
\node [anchor=center,ublue](cb7) at (0.4,-0.1){\Large{$\cdot$}};
\node [anchor=center,ublue](cb8) at (0.35,-0.45){\Large{$\cdot$}};
\node [anchor=center,ublue](cb9) at (0.45,-0.6){\Large{$\cdot$}};
\node [anchor=center,ublue](cb10) at (0.85,-0.45){\Large{$\cdot$}};
\node [anchor=center,ublue](cb11) at (1.45,-0.45){\Large{$\cdot$}};
\node [anchor=center,ublue](cb12) at (1.3,-0.85){\Large{$\cdot$}};
\node [anchor=center,ublue](cb13) at (1.8,-0.5){\Large{$\cdot$}};
\node [anchor=center,ublue](cb14) at (1.75,0.2){\Large{$\cdot$}};
\node [anchor=center,ublue](cb15) at (1.6,0.2){\Large{$\cdot$}};
\end{scope}
%%%%%%%%X、Y词嵌入空间------------------------------------------------------------
\begin{scope}[xshift=-16em]
\draw [-,ublue,line width=0.5pt] (0,0)..controls (0.3,0.2) and (0.5,0)..(0.7,-0.2)..controls (0.8,-0.3) and (0.9,-0.4)..(1.1,-0.4)..controls (1.3,-0.4) and (1.3,-0.1)..(1.28,0)..controls (1.26,0.1) and (1.25,0.2)..(1.2,0.3)..controls (1.15,0.4)and (1.2,0.5)..(1.6,0.55)..controls (1.7,0.56) and (1.78,0.5)..(1.85,0.35)..controls (2.0,0.0) and (2.05,-0.1)..(2.05,-0.5)..controls (2.04,-1.1) and (1.5,-1.1)..(0.6,-0.78)..controls (0.5,-0.74) and (0.4,-0.7)..(0.2,-0.5)..controls(0.1,-0.4) and (-0.15,-0.1)..(0,0) ;
\node [anchor=center](x1) at (-1.45,0.2){\tiny{$\mathbi{X}$}};
\node [anchor=center](y1) at (1.1,0.1){\tiny{$\mathbi{Y}$}};
\node [anchor=center,ublue](cb1) at (0.6,-0.5){\scriptsize{$\bullet$}};
\node [anchor=center,ublue](cb2) at (1.55,-0.8){\scriptsize{$\bullet$}};
\node [anchor=center,ublue](cb3) at (1.6,-0.05){\scriptsize{$\bullet$}};
%%%%%%一堆蓝色的球
\node [anchor=center,ublue](cb4) at (0.1,-0.2){\Large{$\cdot$}};
\node [anchor=center,ublue](cb5) at (0.3,-0.2){\Large{$\cdot$}};
\node [anchor=center,ublue](cb6) at (0.5,-0.25){\Large{$\cdot$}};
\node [anchor=center,ublue](cb7) at (0.4,-0.1){\Large{$\cdot$}};
\node [anchor=center,ublue](cb8) at (0.35,-0.45){\Large{$\cdot$}};
\node [anchor=center,ublue](cb9) at (0.45,-0.6){\Large{$\cdot$}};
\node [anchor=center,ublue](cb10) at (0.85,-0.45){\Large{$\cdot$}};
\node [anchor=center,ublue](cb11) at (1.45,-0.45){\Large{$\cdot$}};
\node [anchor=center,ublue](cb12) at (1.3,-0.85){\Large{$\cdot$}};
\node [anchor=center,ublue](cb13) at (1.8,-0.5){\Large{$\cdot$}};
\node [anchor=center,ublue](cb14) at (1.75,0.2){\Large{$\cdot$}};
\node [anchor=center,ublue](cb15) at (1.6,0.2){\Large{$\cdot$}};
\node [anchor=center](rw1) at (-0.5,0.45){\tiny{cat}};
\node [anchor=center](rw2) at (0.05,0.4){\tiny{feline}};
\node [anchor=center](rw3) at (-1.17,-0.07){\tiny{car}};
\node [anchor=center](rw4) at (-0.7,-0.65){\tiny{deep}};
\node [anchor=center](bw1) at (0.2,-0.1){\tiny{felin}};
\node [anchor=center](bw2) at (0.75,-0.65){\tiny{katze}};
\node [anchor=center](bw3) at (1.55,-0.65){\tiny{auto}};
\node [anchor=center](bw4) at (1.6,-0.2){\tiny{tief}};
\node [anchor=center](de1) at (0.3,-1.5) {\small{(a) $\mathbi{X}$、$\mathbi{Y}$词嵌入空间}};
\node [anchor=center](de2) at (3.9,-1.5) {\small{(b) $\mathbi{X}$映射到$\mathbi{Y}$空间}};
\node [anchor=center](de3) at (7,-1.5) {\small{(c) 词典推断}};
\node [anchor=center](de4) at (10.1,-1.5) {\small{(d) 微调结果}};
\end{scope}
\begin{scope}[xshift=-14.5em,yshift=0.8em,rotate=-150]
\draw [-,red!70,line width=0.5pt] (0.04,-0.5) .. controls (0,-0.4) and (0.4,-0.1)..(0.7,-0.3)..controls (0.9,-0.45) and (1.1,-0.4)..(1.2,-0.3)..controls (1.3,-0.2) and (1.2,0.1).. (1.0,0.3)..controls (0.8,0.5) and (1.0,0.6)..(1.2,0.67)..controls (1.5,0.78) and (1.8,0.5)..(1.9,0.2)..controls(2.1,-0.3) and (2,-0.5)..(1.8,-0.75)..controls (1.5,-1.1) and (1.2,-1.0)..(0.4,-0.8)..controls (0.3,-0.77) and (0.14,-0.755)..(0.04,-0.5);
\node [anchor=center,red!70](cr1) at (0.65,-0.65){\scriptsize{$\bullet$}};
\node [anchor=center,red!70](cr2) at (1.65,-0.65){\scriptsize{$\bullet$}};
\node [anchor=center,red!70](cr3) at (1.5,0.1){\scriptsize{$\bullet$}};
%%%%%%一堆红色的球
\node [anchor=center,red!70](cr4) at (0.15,-0.6){\Large{$\cdot$}};
\node [anchor=center,red!70](cr5) at (0.3,-0.6){\Large{$\cdot$}};
\node [anchor=center,red!70](cr6) at (0.5,-0.55){\Large{$\cdot$}};
\node [anchor=center,red!70](cr7) at (0.35,-0.4){\Large{$\cdot$}};
\node [anchor=center,red!70](cr8) at (0.4,-0.7){\Large{$\cdot$}};
\node [anchor=center,red!70](cr8) at (0.55,-0.8){\Large{$\cdot$}};
\node [anchor=center,red!70](cr9) at (0.9,-0.8){\Large{$\cdot$}};
\node [anchor=center,red!70](cr10) at (0.9,-0.5){\Large{$\cdot$}};
\node [anchor=center,red!70](cr11) at (1.4,-0.8){\Large{$\cdot$}};
\node [anchor=center,red!70](cr12) at (1.45,-0.3){\Large{$\cdot$}};
\node [anchor=center,red!70](cr13) at (1.35,0.3){\Large{$\cdot$}};
\node [anchor=center,red!70](cr14) at (1.2,0.4){\Large{$\cdot$}};
\node [anchor=center,red!70](cr15) at (1.6,0.45){\Large{$\cdot$}};
\end{scope}
%%%%%%%%%%%微调结果------------------------------------------------------------
\begin{scope}[xshift=8.2em]
\draw [-,red!70,line width=0.5pt] (0,0.4688)..controls (0.3,0.45) and (0.5,0.2)..(0.7,-0.25)..controls (0.8,-0.45) and (0.9,-0.4)..(1.1,-0.4)..controls (1.3,-0.42) and (1.3,-0.12)..(1.28,0)..controls (1.26,0.1) and (1.25,0.2)..(1.2,0.3)..controls (1.13,0.4) and (1.18,0.5)..(1.6,0.55)..controls (1.7,0.56) and (1.78,0.5)..(1.85,0.35)..controls (2.03,0.0) and (2.08,-0.1)..(2.07,-0.5)..controls (2.04,-1.1) and (1.5,-1.16)..(0.6,-0.91)..controls (0.05,-0.71) and (-0.2,-0.53)..(-0.25,-0.45)..controls (-0.55,0.0) and (-0.5,0.501)..(0,0.4688);
\draw [-,ublue,line width=0.5pt] (0,0.5)..controls (0.3,0.5) and (0.5,0.2)..(0.7,-0.25)..controls (0.8,-0.45) and (0.9,-0.4)..(1.1,-0.4)..controls (1.3,-0.40) and (1.3,-0.1)..(1.28,0)..controls (1.26,0.1) and (1.25,0.2)..(1.2,0.3)..controls (1.15,0.4)and (1.2,0.5)..(1.6,0.55)..controls (1.7,0.56) and (1.78,0.5)..(1.85,0.35)..controls (2.0,0.0) and (2.05,-0.1)..(2.05,-0.5)..controls (2.04,-1.1) and (1.5,-1.1)..(0.6,-0.91)..controls (0.0,-0.75) and (-0.2,-0.53)..(-0.25,-0.45)..controls (-0.5,0.0) and (-0.5,0.501)..(0,0.5);
\draw [-,thick] (-0.8,1.0)--(-0.8,-1.0);
\node [anchor=center](c1) at (0.1,0.6){\tiny{$\mathbi{Y}$}};
\node [anchor=center](c2) at (-0.45,-0.7){\tiny{$\mathbi{W}\cdot \mathbi{X}$}};
\node [anchor=center,red!70](cr1) at (0.2,-0.35){\scriptsize{$\bullet$}};
\node [anchor=center,red!70](cr2) at (1.58,-0.78){\scriptsize{$\bullet$}};
\node [anchor=center,red!70](cr3) at (1.6,0){\scriptsize{$\bullet$}};
\node [anchor=center,ublue](cb1) at (0.2,-0.3){\scriptsize{$\bullet$}};
\node [anchor=center,ublue](cb2) at (1.55,-0.8){\scriptsize{$\bullet$}};
\node [anchor=center,ublue](cb3) at (1.6,-0.05){\scriptsize{$\bullet$}};
%%%%%%一堆红色的球
\node [anchor=center,red!70](cb4) at (-0.35,0.16){\Large{$\cdot$}};
\node [anchor=center,red!70](cb5) at (-0.03,0.37){\Large{$\cdot$}};
\node [anchor=center,red!70](cb6) at (-0.03,0.12){\Large{$\cdot$}};
\node [anchor=center,red!70](cb7) at (0.37,0.02){\Large{$\cdot$}};
\node [anchor=center,red!70](cb8) at (-0.18,-0.18){\Large{$\cdot$}};
\node [anchor=center,red!70](cb9) at (0.65,-0.43){\Large{$\cdot$}};
\node [anchor=center,red!70](cb10) at (0.32,-0.68){\Large{$\cdot$}};
\node [anchor=center,red!70](cb11) at (0.82,-0.73){\Large{$\cdot$}};
\node [anchor=center,red!70](cb12) at (1.23,-0.85){\Large{$\cdot$}};
\node [anchor=center,red!70](cb13) at (1.8,-0.47){\Large{$\cdot$}};
\node [anchor=center,red!70](cb14) at (1.75,0.23){\Large{$\cdot$}};
\node [anchor=center,red!70](cb15) at (1.38,-0.44){\Large{$\cdot$}};
\node [anchor=center,red!70](cb16) at (1.42,0.26){\Large{$\cdot$}};
%%%%%%一堆蓝色的球
\node [anchor=center,ublue](cb4) at (-0.35,0.2){\Large{$\cdot$}};
\node [anchor=center,ublue](cb5) at (0,0.4){\Large{$\cdot$}};
\node [anchor=center,ublue](cb6) at (0,0.15){\Large{$\cdot$}};
\node [anchor=center,ublue](cb7) at (0.4,0.05){\Large{$\cdot$}};
\node [anchor=center,ublue](cb8) at (-0.15,-0.15){\Large{$\cdot$}};
\node [anchor=center,ublue](cb9) at (0.65,-0.4){\Large{$\cdot$}};
\node [anchor=center,ublue](cb10) at (0.3,-0.65){\Large{$\cdot$}};
\node [anchor=center,ublue](cb11) at (0.8,-0.7){\Large{$\cdot$}};
\node [anchor=center,ublue](cb12) at (1.2,-0.85){\Large{$\cdot$}};
\node [anchor=center,ublue](cb13) at (1.8,-0.5){\Large{$\cdot$}};
\node [anchor=center,ublue](cb14) at (1.75,0.2){\Large{$\cdot$}};
\node [anchor=center,ublue](cb15) at (1.4,-0.45){\Large{$\cdot$}};
\node [anchor=center,ublue](cb16) at (1.45,0.3){\Large{$\cdot$}};
\node [anchor=center](rw1) at (0.22,-0.45){\tiny{cat}};
\node [anchor=center](rw2) at (0.20,-0.15){\tiny{katze}};
\end{scope}
\end{tikzpicture}
\ No newline at end of file
Chapter16/chapter16.aux deleted 100644 → 0
\relax
\providecommand\zref@newlabel[2]{}
\providecommand\hyper@newdestlabel[2]{}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {chapter}{\numberline {1}低资源神经机器翻译}{11}{chapter.1}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\addvspace {10\p@ }}
\@writefile{lot}{\defcounter {refsection}{0}\relax }\@writefile{lot}{\addvspace {10\p@ }}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {section}{\numberline {1.1}数据的有效使用}{11}{section.1.1}\protected@file@percent }
\newlabel{effective-use-of-data}{{1.1}{11}{数据的有效使用}{section.1.1}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.1.1}数据增强}{12}{subsection.1.1.1}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{1. 回译}{12}{section*.3}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.1}{\ignorespaces \color {red}{回译方法的流程(新)} {\color {blue} 图比以前清晰了，但是还是有些乱，可能你陷入到固有思维里了，可以找我再讨论下！}\relax }}{12}{figure.caption.4}\protected@file@percent }
\providecommand*\caption@xref[2]{\@setref\relax\@undefined{#1}}
\newlabel{fig:16-1-xc}{{1.1}{12}{\red {回译方法的流程(新)} {\color {blue} 图比以前清晰了，但是还是有些乱，可能你陷入到固有思维里了，可以找我再讨论下！}\relax }{figure.caption.4}{}}
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.2}{\ignorespaces \color {red}{迭代式回译方法的流程，未修改} {\color {blue} 这个图的逻辑我觉得是ok的，主要是这些线和过程需要再清晰一下，再找我讨论下！}\relax }}{13}{figure.caption.5}\protected@file@percent }
\newlabel{fig:16-2-xc}{{1.2}{13}{\red {迭代式回译方法的流程，未修改} {\color {blue} 这个图的逻辑我觉得是ok的，主要是这些线和过程需要再清晰一下，再找我讨论下！}\relax }{figure.caption.5}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{2. 修改双语数据}{14}{section*.6}\protected@file@percent }
\newlabel{add-noise}{{1.1.1}{14}{2. 修改双语数据}{section*.6}{}}
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.3}{\ignorespaces 三种加噪方法\relax }}{15}{figure.caption.7}\protected@file@percent }
\newlabel{fig:16-4-xc}{{1.3}{15}{三种加噪方法\relax }{figure.caption.7}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{3. 双语句对挖掘}{16}{section*.8}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.4}{\ignorespaces 维基百科中的可比语料\relax }}{17}{figure.caption.9}\protected@file@percent }
\newlabel{fig:16-5-xc}{{1.4}{17}{维基百科中的可比语料\relax }{figure.caption.9}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.1.2}基于语言模型的方法}{17}{subsection.1.1.2}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{1. 语言模型在目标端的融合}{18}{section*.10}\protected@file@percent }
\newlabel{eq:16-1-xc}{{1.1}{18}{1. 语言模型在目标端的融合}{equation.1.1.1}{}}
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.5}{\ignorespaces \color {red}{语言模型的浅融合与深融合，未修改} {\color {blue} 图可以考虑删除了，要不也增加阅读的负担！}\relax }}{18}{figure.caption.11}\protected@file@percent }
\newlabel{fig:16-6-xc}{{1.5}{18}{\red {语言模型的浅融合与深融合，未修改} {\color {blue} 图可以考虑删除了，要不也增加阅读的负担！}\relax }{figure.caption.11}{}}
\newlabel{eq:16-2-xc}{{1.2}{18}{1. 语言模型在目标端的融合}{equation.1.1.2}{}}
\newlabel{eq:16-3-xc}{{1.3}{19}{1. 语言模型在目标端的融合}{equation.1.1.3}{}}
\newlabel{eq:16-4-xc}{{1.4}{19}{1. 语言模型在目标端的融合}{equation.1.1.4}{}}
\newlabel{eq:16-5-xc}{{1.5}{19}{1. 语言模型在目标端的融合}{equation.1.1.5}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{2. 预训练词嵌入}{19}{section*.12}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{3. 预训练模型}{21}{section*.13}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.6}{\ignorespaces \color {red}{MASS 预训练方法，重画}\relax }}{22}{figure.caption.14}\protected@file@percent }
\newlabel{fig:16-8-xc}{{1.6}{22}{\red {MASS 预训练方法，重画}\relax }{figure.caption.14}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{4. 多任务学习}{23}{section*.15}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.7}{\ignorespaces \color {red}{机器翻译中的多任务学习，重画}\relax }}{24}{figure.caption.16}\protected@file@percent }
\newlabel{fig:16-9-xc}{{1.7}{24}{\red {机器翻译中的多任务学习，重画}\relax }{figure.caption.16}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {section}{\numberline {1.2}双向翻译模型}{24}{section.1.2}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.2.1}双向训练}{24}{subsection.1.2.1}\protected@file@percent }
\newlabel{eq:16-6-xc}{{1.6}{24}{双向训练}{equation.1.2.6}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.2.2}对偶学习}{25}{subsection.1.2.2}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{1. 有监督对偶学习}{25}{section*.18}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.8}{\ignorespaces 双向训练的迭代过程\relax }}{26}{figure.caption.17}\protected@file@percent }
\newlabel{fig:16-1-fk}{{1.8}{26}{双向训练的迭代过程\relax }{figure.caption.17}{}}
\newlabel{eq:16-7-xc}{{1.7}{26}{1. 有监督对偶学习}{equation.1.2.7}{}}
\newlabel{eq:16-8-xc}{{1.8}{26}{1. 有监督对偶学习}{equation.1.2.8}{}}
\newlabel{eq:16-2-fk}{{1.9}{27}{1. 有监督对偶学习}{equation.1.2.9}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{2. 无监督对偶学习}{27}{section*.19}\protected@file@percent }
\newlabel{eq:16-9-xc}{{1.10}{27}{2. 无监督对偶学习}{equation.1.2.10}{}}
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.9}{\ignorespaces 无监督对偶学习流程\relax }}{28}{figure.caption.20}\protected@file@percent }
\newlabel{fig:16-10-xc}{{1.9}{28}{无监督对偶学习流程\relax }{figure.caption.20}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {section}{\numberline {1.3}多语言翻译模型}{28}{section.1.3}\protected@file@percent }
\newlabel{multilingual-translation-model}{{1.3}{28}{多语言翻译模型}{section.1.3}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.3.1}基于枢轴语言的方法}{29}{subsection.1.3.1}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.10}{\ignorespaces 基于枢轴语言的翻译过程\relax }}{29}{figure.caption.21}\protected@file@percent }
\newlabel{fig:16-1-ll}{{1.10}{29}{基于枢轴语言的翻译过程\relax }{figure.caption.21}{}}
\newlabel{eq:ll-1}{{1.11}{29}{基于枢轴语言的方法}{equation.1.3.11}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.3.2}基于知识蒸馏的方法}{30}{subsection.1.3.2}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.11}{\ignorespaces 基于知识蒸馏的翻译过程\relax }}{30}{figure.caption.22}\protected@file@percent }
\newlabel{fig:16-2-ll}{{1.11}{30}{基于知识蒸馏的翻译过程\relax }{figure.caption.22}{}}
\newlabel{eq:ll-2}{{1.12}{30}{基于知识蒸馏的方法}{equation.1.3.12}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.3.3}基于迁移学习的方法}{31}{subsection.1.3.3}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.12}{\ignorespaces 传统机器学习\&迁移学习对比\relax }}{31}{figure.caption.23}\protected@file@percent }
\newlabel{fig:16-3-ll}{{1.12}{31}{传统机器学习\&迁移学习对比\relax }{figure.caption.23}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{1. 参数初始化方法}{32}{section*.24}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.13}{\ignorespaces 参数初始化方法图\relax }}{32}{figure.caption.25}\protected@file@percent }
\newlabel{fig:16-4-ll}{{1.13}{32}{参数初始化方法图\relax }{figure.caption.25}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{2. 多语言单模型系统}{32}{section*.26}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.14}{\ignorespaces 参数初始化方法图\relax }}{33}{figure.caption.27}\protected@file@percent }
\newlabel{fig:16-5-ll}{{1.14}{33}{参数初始化方法图\relax }{figure.caption.27}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{3. 零资源翻译}{33}{section*.28}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {section}{\numberline {1.4}无监督机器翻译}{34}{section.1.4}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.4.1}无监督词典归纳}{35}{subsection.1.4.1}\protected@file@percent }
\newlabel{unsupervised-dictionary-induction}{{1.4.1}{35}{无监督词典归纳}{subsection.1.4.1}{}}
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.15}{\ignorespaces 词典归纳原理图\relax }}{35}{figure.caption.29}\protected@file@percent }
\newlabel{fig:16-1-lyf}{{1.15}{35}{词典归纳原理图\relax }{figure.caption.29}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{1. 方法框架}{35}{section*.30}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.16}{\ignorespaces 无监督词典归纳流程图（{\color {red} A->a}）\textsuperscript {\textsuperscript {\cite {DBLP:conf/iclr/LampleCRDJ18}}}\relax }}{36}{figure.caption.31}\protected@file@percent }
\newlabel{fig:16-2-lyf}{{1.16}{36}{无监督词典归纳流程图（{\color {red} A->a}）\upcite {DBLP:conf/iclr/LampleCRDJ18}\relax }{figure.caption.31}{}}
\newlabel{eq:16-1}{{1.14}{37}{1. 方法框架}{equation.1.4.13}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{2. 鲁棒性问题}{37}{section*.32}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.4.2}无监督统计机器翻译}{38}{subsection.1.4.2}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{1. 无监督短语归纳}{38}{section*.33}\protected@file@percent }
\newlabel{eq:16-2}{{1.15}{38}{1. 无监督短语归纳}{equation.1.4.15}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{2. 无监督权重调优}{39}{section*.34}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.4.3}无监督神经机器翻译}{39}{subsection.1.4.3}\protected@file@percent }
\newlabel{unsupervised-NMT}{{1.4.3}{39}{无监督神经机器翻译}{subsection.1.4.3}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{1. 基于无监督统计机器翻译的方法}{39}{section*.35}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.17}{\ignorespaces 用无监督统计机器翻译训练神经机器翻译\relax }}{40}{figure.caption.36}\protected@file@percent }
\newlabel{fig:16-1}{{1.17}{40}{用无监督统计机器翻译训练神经机器翻译\relax }{figure.caption.36}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{2. 基于无监督词典归纳的方法}{40}{section*.37}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{3. 更深层的融合}{40}{section*.39}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.18}{\ignorespaces 基于无监督词典归纳的方法\relax }}{41}{figure.caption.38}\protected@file@percent }
\newlabel{fig:16-2}{{1.18}{41}{基于无监督词典归纳的方法\relax }{figure.caption.38}{}}
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.19}{\ignorespaces 模型初始化方法的优化\relax }}{41}{figure.caption.40}\protected@file@percent }
\newlabel{fig:16-3}{{1.19}{41}{模型初始化方法的优化\relax }{figure.caption.40}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{4. 其它问题}{41}{section*.41}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.20}{\ignorespaces 无监督神经机器翻译模型训练流程\relax }}{43}{figure.caption.42}\protected@file@percent }
\newlabel{fig:16-4}{{1.20}{43}{无监督神经机器翻译模型训练流程\relax }{figure.caption.42}{}}
\@writefile{lot}{\defcounter {refsection}{0}\relax }\@writefile{lot}{\contentsline {table}{\numberline {1.1}{\ignorespaces 三种噪声函数（原句为``我\ 喜欢\ 吃\ 苹果\ 。''）。\relax }}{44}{table.caption.43}\protected@file@percent }
\newlabel{tab:16-1}{{1.1}{44}{三种噪声函数（原句为``我\ 喜欢\ 吃\ 苹果\ 。''）。\relax }{table.caption.43}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {section}{\numberline {1.5}领域适应}{44}{section.1.5}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.21}{\ignorespaces 单词pitch（图里标红）在不同领域的不同词义实例\relax }}{44}{figure.caption.44}\protected@file@percent }
\newlabel{fig:16-1-wbh}{{1.21}{44}{单词pitch（图里标红）在不同领域的不同词义实例\relax }{figure.caption.44}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.5.1}统计机器翻译中的领域适应}{45}{subsection.1.5.1}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{1. 基于混合模型的方法}{45}{section*.45}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{2. 基于数据加权的方法}{45}{section*.46}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{3. 基于数据选择的方法}{46}{section*.47}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{4. 基于伪数据的方法}{46}{section*.48}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.5.2}基于数据的神经机器翻译领域适应}{46}{subsection.1.5.2}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{1. 基于多领域数据的方法}{46}{section*.49}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{2. 基于数据选择的方法}{47}{section*.50}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{3. 基于单语数据的方法}{47}{section*.51}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsection}{\numberline {1.5.3}基于模型的神经机器翻译领域适应}{48}{subsection.1.5.3}\protected@file@percent }
\newlabel{modeling-methods-in neural-machine-translation}{{1.5.3}{48}{基于模型的神经机器翻译领域适应}{subsection.1.5.3}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{1. 基于模型结构的方法}{48}{section*.52}\protected@file@percent }
\@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\contentsline {figure}{\numberline {1.22}{\ignorespaces 领域判别器示意图\relax }}{48}{figure.caption.53}\protected@file@percent }
\newlabel{fig:16-2-wbh}{{1.22}{48}{领域判别器示意图\relax }{figure.caption.53}{}}
\newlabel{eq:16-1-wbh}{{1.16}{48}{1. 基于模型结构的方法}{equation.1.5.16}{}}
\newlabel{eq:16-2-wbh}{{1.17}{48}{1. 基于模型结构的方法}{equation.1.5.17}{}}
\newlabel{eq:16-3-wbh}{{1.18}{49}{1. 基于模型结构的方法}{equation.1.5.18}{}}
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{2. 基于训练策略的方法}{49}{section*.54}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {subsubsection}{3. 基于模型推断的方法}{50}{section*.55}\protected@file@percent }
\@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\contentsline {section}{\numberline {1.6}小结及扩展阅读}{50}{section.1.6}\protected@file@percent }
\@setckpt{Chapter16/chapter16}{
\setcounter{page}{52}
\setcounter{equation}{18}
\setcounter{enumi}{0}
\setcounter{enumii}{0}
\setcounter{enumiii}{0}
\setcounter{enumiv}{0}
\setcounter{footnote}{0}
\setcounter{mpfootnote}{0}
\setcounter{part}{0}
\setcounter{chapter}{1}
\setcounter{section}{6}
\setcounter{subsection}{0}
\setcounter{subsubsection}{0}
\setcounter{paragraph}{0}
\setcounter{subparagraph}{0}
\setcounter{figure}{22}
\setcounter{table}{1}
\setcounter{tabx@nest}{0}
\setcounter{listtotal}{0}
\setcounter{listcount}{0}
\setcounter{liststart}{0}
\setcounter{liststop}{0}
\setcounter{citecount}{0}
\setcounter{citetotal}{0}
\setcounter{multicitecount}{0}
\setcounter{multicitetotal}{0}
\setcounter{instcount}{348}
\setcounter{maxnames}{3}
\setcounter{minnames}{1}
\setcounter{maxitems}{3}
\setcounter{minitems}{1}
\setcounter{citecounter}{0}
\setcounter{maxcitecounter}{0}
\setcounter{savedcitecounter}{0}
\setcounter{uniquelist}{0}
\setcounter{uniquename}{0}
\setcounter{refsection}{0}
\setcounter{refsegment}{0}
\setcounter{maxextratitle}{0}
\setcounter{maxextratitleyear}{0}
\setcounter{maxextraname}{10}
\setcounter{maxextradate}{0}
\setcounter{maxextraalpha}{0}
\setcounter{abbrvpenalty}{50}
\setcounter{highnamepenalty}{50}
\setcounter{lownamepenalty}{25}
\setcounter{maxparens}{3}
\setcounter{parenlevel}{0}
\setcounter{mincomprange}{10}
\setcounter{maxcomprange}{100000}
\setcounter{mincompwidth}{1}
\setcounter{afterword}{0}
\setcounter{savedafterword}{0}
\setcounter{annotator}{0}
\setcounter{savedannotator}{0}
\setcounter{author}{0}
\setcounter{savedauthor}{0}
\setcounter{bookauthor}{0}
\setcounter{savedbookauthor}{0}
\setcounter{commentator}{0}
\setcounter{savedcommentator}{0}
\setcounter{editor}{0}
\setcounter{savededitor}{0}
\setcounter{editora}{0}
\setcounter{savededitora}{0}
\setcounter{editorb}{0}
\setcounter{savededitorb}{0}
\setcounter{editorc}{0}
\setcounter{savededitorc}{0}
\setcounter{foreword}{0}
\setcounter{savedforeword}{0}
\setcounter{holder}{0}
\setcounter{savedholder}{0}
\setcounter{introduction}{0}
\setcounter{savedintroduction}{0}
\setcounter{namea}{0}
\setcounter{savednamea}{0}
\setcounter{nameb}{0}
\setcounter{savednameb}{0}
\setcounter{namec}{0}
\setcounter{savednamec}{0}
\setcounter{translator}{0}
\setcounter{savedtranslator}{0}
\setcounter{shortauthor}{0}
\setcounter{savedshortauthor}{0}
\setcounter{shorteditor}{0}
\setcounter{savedshorteditor}{0}
\setcounter{labelname}{0}
\setcounter{savedlabelname}{0}
\setcounter{institution}{0}
\setcounter{savedinstitution}{0}
\setcounter{lista}{0}
\setcounter{savedlista}{0}
\setcounter{listb}{0}
\setcounter{savedlistb}{0}
\setcounter{listc}{0}
\setcounter{savedlistc}{0}
\setcounter{listd}{0}
\setcounter{savedlistd}{0}
\setcounter{liste}{0}
\setcounter{savedliste}{0}
\setcounter{listf}{0}
\setcounter{savedlistf}{0}
\setcounter{location}{0}
\setcounter{savedlocation}{0}
\setcounter{organization}{0}
\setcounter{savedorganization}{0}
\setcounter{origlocation}{0}
\setcounter{savedoriglocation}{0}
\setcounter{origpublisher}{0}
\setcounter{savedorigpublisher}{0}
\setcounter{publisher}{0}
\setcounter{savedpublisher}{0}
\setcounter{language}{0}
\setcounter{savedlanguage}{0}
\setcounter{origlanguage}{0}
\setcounter{savedoriglanguage}{0}
\setcounter{pageref}{0}
\setcounter{savedpageref}{0}
\setcounter{textcitecount}{0}
\setcounter{textcitetotal}{0}
\setcounter{textcitemaxnames}{0}
\setcounter{biburlbigbreakpenalty}{100}
\setcounter{biburlbreakpenalty}{200}
\setcounter{biburlnumpenalty}{0}
\setcounter{biburlucpenalty}{0}
\setcounter{biburllcpenalty}{0}
\setcounter{smartand}{1}
\setcounter{bbx:relatedcount}{0}
\setcounter{bbx:relatedtotal}{0}
\setcounter{parentequation}{0}
\setcounter{notation}{0}
\setcounter{dummy}{0}
\setcounter{problem}{0}
\setcounter{exerciseT}{0}
\setcounter{exampleT}{0}
\setcounter{vocabulary}{0}
\setcounter{definitionT}{0}
\setcounter{mdf@globalstyle@cnt}{0}
\setcounter{mdfcountframes}{0}
\setcounter{mdf@env@i}{0}
\setcounter{mdf@env@ii}{0}
\setcounter{mdf@zref@counter}{0}
\setcounter{Item}{0}
\setcounter{Hfootnote}{0}
\setcounter{Hy@AnnotLevel}{0}
\setcounter{bookmark@seq@number}{0}
\setcounter{caption@flags}{0}
\setcounter{continuedfloat}{0}
\setcounter{cp@cnt}{0}
\setcounter{cp@tempcnt}{0}
\setcounter{subfigure}{0}
\setcounter{lofdepth}{1}
\setcounter{subtable}{0}
\setcounter{lotdepth}{1}
\setcounter{@pps}{0}
\setcounter{@ppsavesec}{0}
\setcounter{@ppsaveapp}{0}
\setcounter{tcbbreakpart}{0}
\setcounter{tcblayer}{0}
\setcounter{tcolorbox@number}{0}
\setcounter{section@level}{1}
}
Chapter16/chapter16.tex
......@@ -53,7 +53,7 @@
\begin{figure}[htp]
\centering
\input{./Chapter16/Figures/figure-application-process-of-back-translation}
\caption{\red{回译方法的流程(新)} {\color{blue} 图比以前清晰了，但是还是有些乱，可能你陷入到固有思维里了，可以找我再讨论下！}}
\caption{回译方法的流程}
\label{fig:16-1-xc}
\end{figure}
%-------------------------------------------
......@@ -563,18 +563,18 @@ $\funp{P}(\seq{y}|\seq{x})$和$\funp{P}(\seq{x}|\seq{y})$是否真的没有关
\end{itemize}
%----------------------------------------------------------------------------------------
% NEW SECTION
% NEW SECTION 16.4
%----------------------------------------------------------------------------------------
\section{无监督机器翻译}
\parinterval 低资源机器翻译的一种极端情况是：没有任何可以用于模型训练的双语平行数据。一种思路是借用多语言翻译方面的技术（XXX节），利用基于中介语或者零样本学习的方法构建翻译系统。但是，这类方法仍然需要多个语种的平行数据。对于某一个语言对，在只有源语言和目标语言单语数据的前提下，是否仍然可以训练一个有效的翻译模型呢？我们称这种不需要双语数据的机器翻译方法为{\small\bfnew{无监督机器翻译}}\index{无监督机器翻译}（Un-supervised Machine Translation\index{Un-supervised Machine Translation}）。
\parinterval 低资源机器翻译的一种极端情况是：没有任何可以用于模型训练的双语平行数据。一种思路是借用多语言翻译方面的技术\ref{{multilingual-translation-model}}，利用基于中介语或者零样本学习的方法构建翻译系统。但是，这类方法仍然需要多个语种的平行数据。对于某一个语言对，在只有源语言和目标语言单语数据的前提下，是否仍然可以训练一个有效的翻译模型呢？我们称这种不需要双语数据的机器翻译方法为{\small\bfnew{无监督机器翻译}}\index{无监督机器翻译}（Un-supervised Machine Translation\index{Un-supervised Machine Translation}）。
\parinterval 直接进行无监督机器翻译是困难的。一个简单可行的思路是先把问题进行分解，然后分别解决各个子问题，最后形成完整的解决方案。放到无监督机器翻译里面，可以首先使用无监督方法寻找词与词之间的翻译，然后在这基础上，进一步得到句子到句子的翻译模型。
%----------------------------------------------------------------------------------------
% NEW SUB-SECTION
% NEW SUB-SECTION 16.4.1
%----------------------------------------------------------------------------------------
\subsection{无监督词典归纳}\label{unsupervised-dictionary-induction}
......@@ -610,7 +610,7 @@ $\funp{P}(\seq{y}|\seq{x})$和$\funp{P}(\seq{x}|\seq{y})$是否真的没有关
\begin{itemize}
\vspace{0.5em}
\item 对于图\ref{fig:16-2-lyf}(a)中的分布在不同空间中的两个单语词嵌入$\mathbi{x}$和$\mathbi{y}$，基于两者近似同构的假设，利用无监督匹配的方法来得到一个粗糙的线性映射$\mathbi{W}$，使得两个空间能大致对齐，结果如图\ref{fig:16-2-lyf}(b)所示。
\item 对于图\ref{fig:16-2-lyf}(a)中的分布在不同空间中的两个单语词嵌入$\mathbi{X}$和$\mathbi{Y}$，基于两者近似同构的假设，利用无监督匹配的方法来得到一个粗糙的线性映射$\mathbi{W}$，使得两个空间能大致对齐，结果如图\ref{fig:16-2-lyf}(b)所示。
\vspace{0.5em}
\item 在此共享空间中执行对齐算法从而归纳出一个种子词典，如图\ref{fig:16-2-lyf}(c)所示。
\vspace{0.5em}
......@@ -620,8 +620,8 @@ $\funp{P}(\seq{y}|\seq{x})$和$\funp{P}(\seq{x}|\seq{y})$是否真的没有关
\begin{figure}[h]
\centering
\includegraphics[scale=0.6]{Chapter16/Figures/figure-bilingual-dictionary-Induction}
\caption{无监督词典归纳流程图（{\color{red} A->a}）\upcite{DBLP:conf/iclr/LampleCRDJ18}}
\input{Chapter16/Figures/figure-bilingual-dictionary-Induction}
\caption{无监督词典归纳流程图\upcite{DBLP:conf/iclr/LampleCRDJ18}}
\label{fig:16-2-lyf}
\end{figure}
......@@ -629,23 +629,23 @@ $\funp{P}(\seq{y}|\seq{x})$和$\funp{P}(\seq{x}|\seq{y})$是否真的没有关
\begin{itemize}
\vspace{0.5em}
\item 基于GAN的方法\upcite{DBLP:conf/iclr/LampleCRDJ18,DBLP:conf/acl/ZhangLLS17,DBLP:conf/emnlp/XuYOW18,DBLP:conf/naacl/MohiuddinJ19}。在这个任务中，通过生成器来产生映射$\mathbi{W}$，鉴别器负责区分随机抽样的元素$\mathbi{W}\cdot \seq{x}$ 和$\seq{y}$，两者共同优化收敛后即可得到映射$\mathbi{W}$。
\item 基于GAN的方法\upcite{DBLP:conf/iclr/LampleCRDJ18,DBLP:conf/acl/ZhangLLS17,DBLP:conf/emnlp/XuYOW18,DBLP:conf/naacl/MohiuddinJ19}。在这个任务中，通过生成器来产生映射$\mathbi{W}$，鉴别器负责区分随机抽样的元素$\mathbi{W}\cdot \mathbi{X}$ 和$\mathbi{Y}$，两者共同优化收敛后即可得到映射$\mathbi{W}$。
\vspace{0.5em}
\item 基于Gromov-Wasserstein 的方法\upcite{DBLP:conf/emnlp/Alvarez-MelisJ18,DBLP:conf/lrec/GarneauGBDL20,DBLP:journals/corr/abs-1811-01124,DBLP:conf/emnlp/XuYOW18}。Wasserstein距离是度量空间中定义两个概率分布之间距离的函数。在这个任务中，它用来衡量不同语言中单词对之间的相似性，利用空间近似同构的信息可以定义出一些目标函数，之后通过优化该目标函数也可以得到映射$\mathbi{W}$。
\vspace{0.5em}
\end{itemize}
\parinterval 在得到映射$\mathbi{W}$之后，对于$\mathbi{x}$中的任意一个单词$x_{i}$，通过$\mathbi{W}\cdot \mathbi{E}({x}_{i})$将其映射到空间$\mathbi{y}$中（$\mathbi{E}({x}_{i})$表示的是单词$x_{i}$的词嵌入向量），然后在$\mathbi{y}$中找到该点的最近邻点$y_{j}$，于是$y_{j}$就是$x_{i}$的翻译词，重复该过程即可归纳出种子词典$D$，第一阶段结束。事实上，由于第一阶段缺乏监督信号，得到的种子词典$D$会包含大量的噪音，性能并不高，因此需要进行进一步的微调。
\parinterval 在得到映射$\mathbi{W}$之后，对于$\mathbi{X}$中的任意一个单词$x_{i}$，通过$\mathbi{W}\cdot \mathbi{E}({x}_{i})$将其映射到空间$\mathbi{y}$中（$\mathbi{E}({x}_{i})$表示的是单词$x_{i}$的词嵌入向量），然后在$\mathbi{Y}$中找到该点的最近邻点$y_{j}$，于是$y_{j}$就是$x_{i}$的翻译词，重复该过程即可归纳出种子词典$D$，第一阶段结束。事实上，由于第一阶段缺乏监督信号，得到的种子词典$D$会包含大量的噪音，性能并不高，因此需要进行进一步的微调。
\parinterval 微调的原理普遍基于普氏分析\upcite{DBLP:journals/corr/MikolovLS13}。假设现在有一个种子词典$D=\left\{x_{i}, y_{i}\right\}$其中${i \in\{1, n\}}$，和两个单语词嵌入$\mathbi{x}$和$\mathbi{y}$，那么就可以将$D$作为{\small\bfnew{映射锚点}}\index{映射锚点}（Anchor\index{Anchor}）学习一个转移矩阵$\mathbi{W}$，使得$\mathbi{W}\cdot \mathbi{x}$与$\mathbi{y}$这两个空间尽可能相近，此外通过对$\mathbi{W}$施加正交约束可以显著提高能\upcite{DBLP:conf/naacl/XingWLL15}，于是这个优化问题就转变成了{\small\bfnew{普鲁克问题}}\index{普鲁克问题}（Procrustes Problem\index{Procrustes Problem}）\upcite{DBLP:conf/iclr/SmithTHH17}，可以通过{\small\bfnew{奇异值分解}}\index{奇异值分解}（Singular Value Decomposition，SVD\index{Singular Value Decomposition，SVD}）来获得近似解：
\parinterval 微调的原理普遍基于普氏分析\upcite{DBLP:journals/corr/MikolovLS13}。假设现在有一个种子词典$D=\left\{x_{i}, y_{i}\right\}$其中${i \in\{1, n\}}$，和两个单语词嵌入$\mathbi{X}$和$\mathbi{Y}$，那么就可以将$D$作为{\small\bfnew{映射锚点}}\index{映射锚点}（Anchor\index{Anchor}）学习一个转移矩阵$\mathbi{W}$，使得$\mathbi{W}\cdot \mathbi{X}$与$\mathbi{Y}$这两个空间尽可能相近，此外通过对$\mathbi{W}$施加正交约束可以显著提高能\upcite{DBLP:conf/naacl/XingWLL15}，于是这个优化问题就转变成了{\small\bfnew{普鲁克问题}}\index{普鲁克问题}（Procrustes Problem\index{Procrustes Problem}）\upcite{DBLP:conf/iclr/SmithTHH17}，可以通过{\small\bfnew{奇异值分解}}\index{奇异值分解}（Singular Value Decomposition，SVD\index{Singular Value Decomposition，SVD}）来获得近似解：
\begin{eqnarray}
\mathbi{W}^{\star} & = &\underset{\mathbi{W} \in O_{d}(\mathbb{R})}{\operatorname{argmin}}\|\mathbi{W}\cdot \mathbi{x}'- \mathbi{y}' \|_{\mathrm{F}}=\mathbi{U}\cdot \mathbi{V}^{\rm{T}} \\
\textrm{s.t.\ \ \ \ } \mathbi{U} \Sigma \mathbi{V}^{\rm{T}} &= &\operatorname{SVD}\left(\mathbi{y}'\cdot \mathbi{x}'^{\rm{T}}\right)
\mathbi{W}^{\star} & = &\underset{\mathbi{W} \in O_{d}(\mathbb{R})}{\operatorname{argmin}}\|\mathbi{W}\cdot \mathbi{X}'- \mathbi{Y}' \|_{\mathrm{F}}=\mathbi{U}\cdot \mathbi{V}^{\rm{T}} \\
\textrm{s.t.\ \ \ \ } \mathbi{U} \Sigma \mathbi{V}^{\rm{T}} &= &\operatorname{SVD}\left(\mathbi{Y}'\cdot \mathbi{X}'^{\rm{T}}\right)
\label{eq:16-1}
\end{eqnarray}
\noindent 其中， $\operatorname{SVD}(\cdot)$表示奇异值分解，$\mathbi{y}'$和$\mathbi{x}'$中的单词来自$D$且行对齐。利用上式可以获得新的$\mathbi{W}$，通过$\mathbi{W}$可以归纳出新的$D$，如此迭代进行微调最后即可以得到收敛的$D$。
\noindent 其中， $\operatorname{SVD}(\cdot)$表示奇异值分解，$\mathbi{Y}'$和$\mathbi{X}'$中的单词来自$D$且行对齐。利用上式可以获得新的$\mathbi{W}$，通过$\mathbi{W}$可以归纳出新的$D$，如此迭代进行微调最后即可以得到收敛的$D$。
\parinterval 较早的无监督方法是基于GAN\upcite{DBLP:conf/acl/ZhangLLS17,DBLP:conf/emnlp/ZhangLLS17,DBLP:conf/iclr/LampleCRDJ18}，这是一个很自然的想法，利用生成器产生映射然后用判别器来区别两个空间，尽管它取得了不错的效果，然而研究表明GAN缺乏稳定性，容易在低资源语言对上失败\upcite{hartmann2018empirical}，因此有不少改进的工作，比如：利用{\small\bfnew{变分自编码器}}（Variational Autoencoders，VAEs）来捕获更深层次的语义信息并结合对抗训练的方法\upcite{DBLP:conf/emnlp/DouZH18,DBLP:conf/naacl/MohiuddinJ19}；通过改进最近邻点的度量函数来提升性能的方法\upcite{DBLP:conf/acl/HuangQC19,DBLP:conf/emnlp/JoulinBMJG18}；利用多语言信号来提升性能的方法\upcite{DBLP:conf/emnlp/ChenC18,DBLP:conf/emnlp/TaitelbaumCG19,DBLP:journals/corr/abs-1811-01124,DBLP:conf/naacl/HeymanVVM19}；也有一些工作舍弃GAN，通过直接优化度量空间距离来进行匹配的方法\upcite{DBLP:conf/emnlp/HoshenW18,DBLP:conf/emnlp/XuYOW18,DBLP:conf/emnlp/Alvarez-MelisJ18,DBLP:conf/emnlp/MukherjeeYH18}。此外，也有另外一些工作是旨在分析或提升无监督词典归纳的鲁棒性。比如通过大量实验来分析无监督词典归纳任务的局限性、难点以及挑战\upcite{DBLP:conf/acl/SogaardVR18,DBLP:conf/acl/OrmazabalALSA19,DBLP:conf/emnlp/VulicGRK19,DBLP:conf/emnlp/HartmannKS18}；分析和对比目前各种无监督方法的性能\upcite{DBLP:conf/nips/HartmannKS19}；通过实验分析指出目前所用的数据集存在的问题\upcite{DBLP:conf/emnlp/Kementchedjhieva19}。
......@@ -674,7 +674,7 @@ $\funp{P}(\seq{y}|\seq{x})$和$\funp{P}(\seq{x}|\seq{y})$是否真的没有关
%----------------------------------------------------------------------------------------
% NEW SUB-SECTION
% NEW SUB-SECTION 16.4.2
%----------------------------------------------------------------------------------------
\subsection{无监督统计机器翻译}
......@@ -728,13 +728,14 @@ P(\mathbi{y}|\mathbi{x}) & = & \frac{\mathrm{cos}(\mathbi{x},\mathbi{y})/\tau}{\
\subsubsection{1. 基于无监督统计机器翻译的方法}
一个简单的方法是，借助已经成功的无监督方法来为神经机器翻译模型提供少量双语监督信号，然后在这个基础上训练模型。由于初始的监督信号可能很少或者包含大量噪声，因此需要逐步优化数据来重新训练出更好的模型。这也是目前所有无监督神经机器翻译方法的核心思路。这个方案最简单直接的实现就是借助已经成功的无监督统计机器翻译模型产生伪双语数据来训练神经机器翻译模型 ，然后模型进行迭代回译来进行数据优化，如图\ref{fig:16-1} 所示\upcite{DBLP:conf/acl/ArtetxeLA19}。这个方法的优点是直观，并且性能稳定，容易调试（所有模块都互相独立）。缺点是复杂繁琐，涉及许多超参数调整工作，而且训练代价较大。
%-------------------------------
\begin{figure}[h]
\centering
\includegraphics[scale=0.2,angle=90]{Chapter16/Figures/figure-unmt-idea1.jpg}
\caption{用无监督统计机器翻译训练神经机器翻译}
\label{fig:16-1}
\end{figure}
%-------------------------------
%----------------------------------------------------------------------------------------
% NEW SUB-SUB-SECTION
......@@ -753,14 +754,14 @@ P(\mathbi{y}|\mathbi{x}) & = & \frac{\mathrm{cos}(\mathbi{x},\mathbi{y})/\tau}{\
\end{itemize}
而第二阶段的改写任务其实也是一个特殊的翻译任务，只不过现在的源语言和目标语言是使用不同的方式表达的同一种语言的句子。因此可以使用神经机器翻译模型来完成这个任务，而且由于这里不涉及双语数据而只需要单语数据，模型的训练也将是无监督的。这样的方法不再需要无监督统计机器翻译，并且适应能力很强。对于新语种，不需要重新训练神经机器翻译模型，只需要训练无监督词典归纳进行词的翻译，然后使用相同的模型进行改写。但是，目前训练数据需要使用其他语种对的双语数据来进行构造（把源语言句子里每个词使用双语词典进行翻译作为输入，输出的目标语言句子不变）。虽然可以通过把单语句子根据规则或者随机进行打乱来生成训练数据，但是这些句子不符合语言学的规律，并且跟真实句子所服从的数据分布不一致，导致训练测试不一致的问题。而且这样一个两阶段的过程会产生错误传播的问题，比如无监督词典归纳对一些词进行了错误的翻译，那么这些错误的翻译会被送到神经机器翻译模型里进行改写，而因为翻译模型这时候已经无法看到源语言句子来进行修正，所以最终的结果将继承无监督词典归纳的错误\upcite{DBLP:conf/acl/PourdamghaniAGK19}。
%-------------------------------
\begin{figure}[h]
\centering
\includegraphics[scale=0.2,angle=90]{Chapter16/Figures/figure-unmt-idea2.jpg}
\caption{基于无监督词典归纳的方法}
\label{fig:16-2}
\end{figure}
%-------------------------------
%----------------------------------------------------------------------------------------
% NEW SUB-SUB-SECTION
%----------------------------------------------------------------------------------------
......@@ -768,14 +769,14 @@ P(\mathbi{y}|\mathbi{x}) & = & \frac{\mathrm{cos}(\mathbi{x},\mathbi{y})/\tau}{\
\subsubsection{3. 更深层的融合}
\parinterval 为了获得更好的神经机器翻译模型，可以对训练流程和模型做更深度的整合。{\chapternine}已经介绍，神经机器翻译模型的训练包含两个阶段：初始化和优化，而无监督神经机器翻译的核心思路也是对应的两个阶段：无监督方法提供初始的监督信号和数据优化，因此可以考虑通过在模型的初始化阶段使用无监督方法提供初始的监督信号，然后优化过程不但优化模型的参数，还优化训练使用的数据，从而避免流水线带来的错误传播。其中初始的监督信号可以通过两种方法提供给模型，一种是直接使用无监督方法提供最初的伪双语数据来训练最初的翻译模型，另一种则是借助无监督方法来初始化模型，得到最初的翻译模型后直接使用初始化好的翻译模型产生伪双语数据来训练自己，如图\ref{fig:16-3}所示。图\ref{fig:16-3}(a)的一个简单实现是利用无监督词典归纳得到的词典对单语数据进行逐词的翻译，得到最初的伪双语数据，然后在这些数据上训练最初的翻译模型，最后不断地交替优化数据和模型，得到更好的翻译模型和质量更好的伪数据\upcite{DBLP:conf/iclr/LampleCDR18}。这样的做法通过不断优化训练用的双语数据，摆脱了无监督词典归纳在最初的伪双语数据中遗留下来的错误，同时也避免了使用无监督统计机器翻译模型的繁琐和代价。图\ref{fig:16-3}(b)的实现则依赖于具体的翻译模型初始化方法，我们将在下一节讨论翻译模型的不同初始化方法。
%-------------------------------
\begin{figure}[h]
\centering
\input{Chapter16/Figures/figure-optimization-of-the-model-initialization-method}
\caption{模型初始化方法的优化}
\label{fig:16-3}
\end{figure}
%-------------------------------
%----------------------------------------------------------------------------------------
% NEW SUB-SUB-SECTION
%----------------------------------------------------------------------------------------
......@@ -797,14 +798,14 @@ P(\mathbi{y}|\mathbi{x}) & = & \frac{\mathrm{cos}(\mathbi{x},\mathbi{y})/\tau}{\
\end{itemize}
\parinterval 最后图\ref{fig:16-4}简单总结了无监督神经机器翻译的流程。下面分别讨论：无监督神经机器翻译里面模型的初始化，以及语言模型目标函数的选用。
%-------------------------------
\begin{figure}[h]
\centering
\input{Chapter16/Figures/figure-unmt-process}
\caption{无监督神经机器翻译模型训练流程}
\label{fig:16-4}
\end{figure}
%-------------------------------
\begin{itemize}
\vspace{0.5em}
\item {\small\bfnew{模型参数初始化}}。无监督神经机器翻译的关键在于如何提供最开始的监督信号，从而启动后续的迭代流程。无监督词典归纳已经可以提供一些可靠的监督信号，那么如何在模型初始化中融入这些信息？既然神经机器翻译模型都使用词嵌入层作为输入，而无监督词典归纳总是首先把两个语言各自的单语词嵌入映射到一个空间后才归纳双语词典，那么可以使用这些映射后的词嵌入来初始化模型的词嵌入层，然后在这个基础上训练模型，因为这些映射后的词嵌入天然就包含了大量的监督信号，比如，两个语言里意思相近的词对应的词嵌入会比其他词更靠近对方\upcite{DBLP:journals/ipm/FarhanTAJATT20}。 为了防止训练过程中模型参数的更新会破坏词嵌入当中的词对齐信息，通常初始化后会固定模型的词嵌入层不让其更新\upcite{DBLP:conf/emnlp/ArtetxeLA18}。
......@@ -840,14 +841,24 @@ P(\mathbi{y}|\mathbi{x}) & = & \frac{\mathrm{cos}(\mathbi{x},\mathbi{y})/\tau}{\
\section{领域适应}
\parinterval 机器翻译常常面临训练时与应用时所处领域不一致的问题，比如，在一个新闻类数据上训练的系统应用在翻译医学文献上。不同领域的句子通常存在着很大的区别，比如，日常用语的结构较为简单，而化学领域论文的单词和句子结构较为复杂。此外，不同领域之间存在着较为严重的一词多义问题，即同一个词在不同领域中经常会有不同的含义，如图\ref{fig:16-1-wbh}所示。
\parinterval 机器翻译常常面临训练时与应用时所处领域不一致的问题，比如，在一个新闻类数据上训练的系统应用在翻译医学文献上。不同领域的句子通常存在着很大的区别，比如，日常用语的结构较为简单，而化学领域论文的单词和句子结构较为复杂。此外，不同领域之间存在着较为严重的一词多义问题，即同一个词在不同领域中经常会有不同的含义，实例\ref{eg:16-1}展示了单词pitch在不同领域的不同词义。
\begin{figure}[h]
\centering
\includegraphics[scale=3]{Chapter16/Figures/figure-the-meaning-of-pitch-in-different-fields.jpg}
\caption{单词pitch（图里标红）在不同领域的不同词义实例}
\label{fig:16-1-wbh}
\end{figure}
\begin{example}
单词pitch在不同领域的不同词义
体育领域：The rugby tour was a disaster both on and off the {\red{pitch}}.
\qquad\qquad \hspace{0.8em} 这次橄榄球巡回赛在{\red{场}}上、{\red{场}}下都彻底失败。
化学领域：The timbers of similar houses were painted with {\red{pitch}}.
\qquad\qquad \hspace{0.8em} 类似房屋所用的栋木刷了{\red{沥青}}。
声学领域：A basic sense of rhythm and {\red{pitch}} is essential in a music teacher.
\qquad\qquad \hspace{0.8em} 基本的韵律感和{\red{音高感}}是音乐教师的必备素质。
\label{eg:16-1}
\end{example}
\parinterval 在机器翻译任务中，某些领域的双语数据相对容易获取，如新闻等领域，所以机器翻译在这些领域上表现较佳。然而，即使在富资源语种上，化学、医学等专业领域的双语数据却十分有限。如果直接使用低资源领域的数据训练一个机器翻译模型，由于数据稀缺问题，会导致模型的性能较差\upcite{DBLP:conf/iccv/SunSSG17}。混合多个领域的数据进行训练，不同领域的数据量不平衡会导致数据较少的领域训练不充分，模型容易忽略低资源领域的知识，使得在低资源领域上的翻译结果差强人意\upcite{DBLP:conf/acl/DuhNST13}。
......
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论